Corrections
We never silently edit a claim. Every correction gets a new event, a permanent URL, and a visible label — separate from ordinary grade changes (see what changed). A rising correction count is a good sign: it means the re-review system is working.
Policy (COPE-aligned)
| Type | Definition | Response |
|---|---|---|
| Correction (minor) | Typo, broken link, formatting — does not change meaning. | Fixed; noted in page history; no separate correction record. |
| Correction (substantive) | A number, dose, population, citation or qualifier was wrong; the verdict stands. | Permanent /corrections/ URL; page banner 90 days; ledger event. |
| Grade correction | The grade or certainty itself was wrong — not "changed on new evidence," wrong. | Permanent URL; banner; API/dataset/webhook consumers notified; affected AI panel prompts re-run. |
| Safety correction | A safety qualifier, interaction severity, or contraindication was wrong or missing. | 72-hour fast path; stopped serving via API/agents; watchlist alert. |
| Retraction | The claim should never have been published. | Claim moves to retracted; page retained as a tombstone with a full explanation. Never deleted. |
Log
No corrections since launch. This log publishes the moment one exists — we'd rather ship the policy with an honest zero than backfill it later.
Found something wrong? Report an error — one click, no account. 7 grade-change events tracked separately at /changes.